学术报告

题      目：Data-driven studies for inverse problems in imaging

报  告  人：李文彬  副教授  (邀请人：钟柳强 )

                                   哈尔滨工业大学深圳校区

时      间：6月16日  11：00-12：00

地     点：数科院西楼二楼会议室

报告人简介:

        李文彬, 哈尔滨工业大学深圳校区，副教授, 在四川大学先后获得学士和硕士学位, 在香港科技大学获得哲学博士（数学）学位, 导师为梁承裕和王筱平教授, 在 Michigan State University 从事博士后研究工作, 研究领域主要包含反问题建模与计算和偏微分方程数值解.

摘      要：

       We present our recent work in data-driven studies for inverse problems in imaging. The motivation is to deal with large-scale inverse problems with measurement data contaminated by random noises. The following contents will be included. (1) A stochastic gradient descent approach with partitioned-truncated singular value decomposition for large-scale inverse problems of magnetic modulus data. This work aims to develop efficient computational methods for massive datasets in nonlinear magnetic inverse problem which were infeasible to process due to the restriction of computational resource. (2) uniformly convex neural networks and iterated network Tikhonov (iNETT) method. We propose rigorous theories and detailed algorithms for the construction of convex and uniformly convex neural networks. Then we develop an iNETT algorithm which employs data-driven regularizers for the solution of ill-posed inverse problems.